
import pandas as pd 
import numpy as np
import matplotlib.pyplot as plt 
import seaborn as sns 
from collections import defaultdict
from sklearn.preprocessing import LabelEncoder
import joypy
import plotly.express as px 
import matplotlib.pyplot as plt
from sklearn.ensemble import RandomForestClassifier
from sklearn.inspection import permutation_importance
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import confusion_matrix,classification_report, accuracy_score,roc_auc_score,roc_curve
from xgboost import XGBClassifier 
from sklearn.model_selection import RandomizedSearchCV, GridSearchCV
import shap
df = pd.read_excel('Task.xlsx')
df.head()


# the summary of a DataFrame which includes column dtypes, non-null
# we notice that we have no NULL values in the datafrome which save one step in preprocessing
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3333 entries, 0 to 3332
Data columns (total 25 columns):
 #   Column                  Non-Null Count  Dtype  
---  ------                  --------------  -----  
 0   customerID              3333 non-null   object 
 1   gender                  3333 non-null   object 
 2   SeniorCitizen           3333 non-null   int64  
 3   US State                3333 non-null   object 
 4   Duration                3333 non-null   int64  
 5   Phone Code              3333 non-null   int64  
 6   Telephone Number        3333 non-null   object 
 7   Product: International  3333 non-null   object 
 8   Product: Voice mail     3333 non-null   object 
 9   vmail                   3333 non-null   int64  
 10  Call day minutes        3333 non-null   float64
 11  total day calls         3333 non-null   int64  
 12  Total, EUR              3333 non-null   float64
 13  eve minutes             3333 non-null   float64
 14  eve calls               3333 non-null   int64  
 15  eve EUR                 3333 non-null   float64
 16  night minutes           3333 non-null   float64
 17  night calls             3333 non-null   int64  
 18  night EUR               3333 non-null   float64
 19  international minutes   3333 non-null   float64
 20  international calls     3333 non-null   int64  
 21  internatonal EUR        3333 non-null   float64
 22  service calls           3333 non-null   int64  
 23  PaperlessBilling        3333 non-null   object 
 24  churn                   3333 non-null   bool   
dtypes: bool(1), float64(8), int64(9), object(7)
memory usage: 628.3+ KB


# Analyzes the statistical summaries of the numerical columns
df.describe()


# here we can see the number of unique values for each feature 
# we notice that 27 Telephone Numbers are not unique, this will be ignored as the number is very small

df.astype('object').describe().transpose()


# we notice that number of users with no churn is very much higher than the ones with churn
# 85% of users didn't churn in the dataset and only 15% churned 
df["churn"].value_counts()

False    2850
True      483
Name: churn, dtype: int64


#Defining bar chart function
def bar(feature, df=df ):
    #Groupby the categorical feature
    temp_df = df.groupby([feature, 'churn']).size().reset_index()
    temp_df = temp_df.rename(columns={0:'Count'})
    #Calculate the value counts of each distribution and it's corresponding Percentages
    value_counts_df = df[feature].value_counts().to_frame().reset_index()
    categories = [cat[1][0] for cat in value_counts_df.iterrows()]
    #Calculate the value counts of each distribution and it's corresponding Percentages
    num_list = [num[1][1] for num in value_counts_df.iterrows()]
    div_list = [element / sum(num_list) for element in num_list]
    percentage = [round(element * 100,1) for element in div_list]
    #Defining string formatting for graph annotation
    #Numeric section
    def num_format(list_instance):
        formatted_str = ''
        for index,num in enumerate(list_instance):
            if index < len(list_instance)-2:
                formatted_str=formatted_str+f'{num}%, ' #append to empty string(formatted_str)
            elif index == len(list_instance)-2:
                formatted_str=formatted_str+f'{num}% & '
            else:
                formatted_str=formatted_str+f'{num}%'
        return formatted_str
    #Categorical section
    def str_format(list_instance):
        formatted_str = ''
        for index, cat in enumerate(list_instance):
            if index < len(list_instance)-2:
                formatted_str=formatted_str+f'{cat}, '
            elif index == len(list_instance)-2:
                formatted_str=formatted_str+f'{cat} & '
            else:
                formatted_str=formatted_str+f'{cat}'
        return formatted_str


    #Running the formatting functions
    num_str = num_format(percentage)
    cat_str = str_format(categories)
    
    #Setting graph framework
    fig = px.bar(temp_df, x=feature, y='Count', color='churn', title=f'Churn rate by {feature}', barmode="group", color_discrete_sequence=["blue", "red"])
    fig.add_annotation(
                text=f'Value count of distribution of {cat_str} are<br>{num_str} percentage respectively.',
                align='left',
                showarrow=False,
                xref='paper',
                yref='paper',
                x=1.4,
                y=1.3,
                bordercolor='black',
                borderwidth=1)
    fig.update_layout(
        # margin space for the annotations on the right
        margin=dict(r=400),
    )
     
    return fig.show()

# Defining the histogram plotting function
def hist(feature):
    group_df = df.groupby([feature, 'churn']).size().reset_index()
    group_df = group_df.rename(columns={0: 'Count'})
    fig = px.histogram(group_df, x=feature, y='Count', color='churn', marginal='box', title=f'Churn rate frequency to {feature} distribution', color_discrete_sequence=["blue", "red"])
    fig.show()


# converting categorical features to numerical
df.loc[df.PaperlessBilling=='No','PaperlessBilling'] = 0
df.loc[df.PaperlessBilling=='Yes','PaperlessBilling'] = 1

df.loc[df.SeniorCitizen==0,'SeniorCitizen'] =0 
df.loc[df.SeniorCitizen==1,'SeniorCitizen'] =1 

df.loc[df.gender=='Male','gender'] = 0
df.loc[df.gender=='Female','gender'] = 1

df.loc[df['Product: International']=='no','Product: International'] = 0
df.loc[df['Product: International']=='yes','Product: International'] = 1

df.loc[df['Product: Voice mail']=='no','Product: Voice mail'] = 0
df.loc[df['Product: Voice mail']=='yes','Product: Voice mail'] = 1


#printing the distribution for categorical features 

# we notice that mostly the gender had a very small effect on churn which is supported by the correlation
bar('gender')
# we also notice that SeniorCitizen has less effect on churn  which is supported by the correlation
bar('SeniorCitizen')
# it seems that Product International has a great effect on churn  which is supported by the correlation
bar('Product: International')
bar('Product: Voice mail')
# PaperlessBilling seems to have less effect on churn  which is supported by the correlation
bar('PaperlessBilling')
bar('US State')
bar('Phone Code')


# Bivariate analysis: looks at how two variables relate to each other in terms of churn
# Numerical-numerical variables
sns.pairplot(data = df, hue='churn')
plt.show()


# most of the numerical features are normally distributed except vmail, international calls, service calls
# since we're using algorithms based on decision trees, it means that you don't have to worry about "non-normality" completely.

hist('Duration')
hist('vmail')
hist('Call day minutes')
hist('total day calls')
hist('Total, EUR')
hist('eve minutes')
hist('eve calls')
hist('eve EUR')
hist('night minutes')
hist('night calls')
hist('night EUR')
hist('international minutes')
hist('international calls')
hist('internatonal EUR')
hist('service calls')


# extracting the features needed for the task
df_dis_calls=df[['international minutes','eve calls','total day calls','US State']]
df_dis_calls


# generating the plot with no geographical feature
plt.figure(figsize=(13,10), dpi= 80)
sns.distplot(df_dis_calls['international minutes'], color="dodgerblue", label="international minutes", hist_kws={'alpha':.3}, kde_kws={'linewidth':4})
sns.distplot(df_dis_calls['eve calls'], color="red", label="eve calls", hist_kws={'alpha':.3}, kde_kws={'linewidth':4})
sns.distplot(df_dis_calls['total day calls'], color="g", label="total day calls", hist_kws={'alpha':.2}, kde_kws={'linewidth':4})
plt.ylim(0, 0.35)

# Decoration
plt.title('Distribution of international calls minutes in breakdown for evening and day calls', fontsize=22)
plt.legend()
plt.show()

C:\Users\ElHarameinSamyJaby\Anaconda3\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

C:\Users\ElHarameinSamyJaby\Anaconda3\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).

C:\Users\ElHarameinSamyJaby\Anaconda3\lib\site-packages\seaborn\distributions.py:2619: FutureWarning:

`distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).


# generating the plot with geographical feature for each state
# Draw Plot
plt.figure(figsize=(80,100), dpi= 2000)
fig, axes = joypy.joyplot(df_dis_calls, column=['international minutes','eve calls','total day calls'], by='US State', ylim='own', figsize=(50,100), alpha=0.5)

# Decoration
#plt.rc("font", size=50)
plt.title('Per state distribution of international calls minutes in breakdown for evening and day calls', fontsize=100)
plt.xlabel('State', fontsize=120,  color='grey')

plt.show()

<Figure size 160000x200000 with 0 Axes>


# some of the assumptions are already made in step 1

# as seen in the following correlation chart, the most correlated features with churn are the following:
# Product: International
# service calls
# service calls rate
# Total, EUR
# Call day minutes
# Total, EUR rate
# Call day minutes rate
# churn rate

# Most relevant features for churn are usually users engagement with the product and services
# Missing features include product characteristics, for example product type, product price, contract minimum term length,  number of parallel contracts.

# Service related like broadband speed, network coverage
# User type: Private or business, new user flag.


#calculate features by state
df_bystate=df.groupby(['US State'])['total day calls','night calls','international calls','Duration','Total, EUR','service calls','Call day minutes', 'night minutes'].mean().reset_index()
#calculate churn rate by state
df_bystate_churn=df.groupby(['US State']).agg({'churn':'sum', 
                         'customerID':'nunique'}).reset_index()
df_bystate_churn['churn']=df_bystate_churn['churn']/df_bystate_churn['customerID']
del df_bystate_churn['customerID']
df_bystate=pd.merge(df_bystate,
                 df_bystate_churn,
                 on='US State', 
                 how='outer')
df_bystate.rename(columns={'total day calls':'total day calls rate', 'night calls':'night calls rate','international calls':'international calls rate','Duration':'Duration rate','Total, EUR':'Total, EUR rate','service calls':'service calls rate','Call day minutes':'Call day minutes rate','night minutes':'night minutes rate', 'churn':'churn rate'}, inplace=True)
df_bystate

C:\Users\ElHarameinSamyJaby\AppData\Local\Temp\ipykernel_19160\2159489575.py:2: FutureWarning:

Indexing with multiple keys (implicitly converted to a tuple of keys) will be deprecated, use a list instead.


df=pd.merge(df, df_bystate, on='US State', how='outer')
#calculating rates from deviding each feature with same feature per state rate
df['total day calls rate']=df['total day calls']/df['total day calls rate']
df['night calls rate']=df['night calls']/df['night calls rate']
df['international calls rate']=df['international calls']/df['international calls rate']
df['Duration rate']=df['Duration']/df['Duration rate']
df['Total, EUR rate']=df['Total, EUR']/df['Total, EUR rate']
df['service calls rate']=df['service calls']/df['service calls rate']
df['Call day minutes rate']=df['Call day minutes']/df['Call day minutes rate']
df['night minutes rate']=df['night minutes']/df['night minutes rate']
# generate new feature with area code from Telephone Number
df['AreaCode'] = df['Telephone Number'].str[:3]
df


hist('international calls rate')


# Encoding the variable
d = defaultdict(LabelEncoder)
df = df.apply(lambda x: d[x.name].fit_transform(x))
df


# Correlation matrix graph of the data set
f, ax = plt.subplots(figsize= [20,15])
sns.heatmap(df.corr(), annot=True, fmt=".2f", ax=ax)
ax.set_title("Correlation Matrix for churn prediction", fontsize=20)
plt.show()


# calculates the correlation of churn with other features:
plt.figure(figsize=(15,8))
df.corr()['churn'].sort_values(ascending = False).plot(kind='bar')

<AxesSubplot:>


X = df.loc[:, ((df.columns != 'churn') & (df.columns != 'customerID'))]

#for normalizing the range data that has a wide ranging set of independent variables-
churn_scaler = MinMaxScaler()
churn_scaler.fit(X)
X=churn_scaler.transform(X)
X=pd.DataFrame(X, columns = df.loc[:, ((df.columns != 'churn') & (df.columns != 'customerID'))].columns)
y = df['churn']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=12)


# ML models

# Churn prediction here is treated it as a classification problem, classifying a customer as yes/no for churning.


# Logistic Regression: 
# -easy to explain and implement
# -with only thousands of data points, there may not be enough data to get very good results

# Random Forest:
#-can yield good results with less data, so it’s one of the best classification models for churn prediction
#-since we use grid-search method and the feature size is not so big, feature selection isn’t necessary.
#-It will automatically select the best features at each node of ensemble trees anyways.
#-to select the best hyperparameters, a 5 fold cross validation is used 
# searching across 200 different combinations


# XGBoost:
#-good option for mid-size datasets
#-To select the best hyperparameters, a 5-Fold Cross Validation with bayesian search is used.


# Random Forest:

# Feature importance in Random Forest 
plt.rcParams.update({'figure.figsize': (12.0, 8.0)})
plt.rcParams.update({'font.size': 14})


rf = RandomForestClassifier(n_estimators=100, oob_score = True)
rf.fit(X_train, y_train)
sorted_idx = rf.feature_importances_.argsort()
plt.barh(X_train.columns[sorted_idx], rf.feature_importances_[sorted_idx])
plt.xlabel("Random Forest Feature Importance")

Text(0.5, 0, 'Random Forest Feature Importance')


# Finding best hyperparameters wule applying cross validation
# RandomizedSearchCV randomly passes the set of hyperparameters and calculate the score and gives the best set of hyperparameters which gives the best score as an output.

# Number of trees in random forest
n_estimators = [int(x) for x in np.linspace(start = 200, stop = 2000, num = 10)]
# Number of features to consider at every split
max_features = ['auto', 'sqrt']
# Maximum number of levels in tree
max_depth = [int(x) for x in np.linspace(10, 110, num = 11)]
max_depth.append(None)
# Minimum number of samples required to split a node
min_samples_split = [2, 5, 10]
# Minimum number of samples required at each leaf node
min_samples_leaf = [1, 2, 4]
# Method of selecting samples for training each tree
bootstrap = [True, False]

# Create the random grid
random_grid = {'n_estimators': n_estimators,
               'max_features': max_features,
               'max_depth': max_depth,
               'min_samples_split': min_samples_split,
               'min_samples_leaf': min_samples_leaf,
               'bootstrap': bootstrap}

# Use the random grid to search for best hyperparameters
# First create the base model to tune
rf = RandomForestClassifier(random_state = 42)
# Random search of parameters, using 5 fold cross validation, 
# search across 200 different combinations, and use all available cores
rf_random = RandomizedSearchCV(estimator=rf, param_distributions=random_grid,
                              n_iter = 200, scoring='precision', 
                              cv = 5, verbose=2, random_state=42, n_jobs=-1,
                              return_train_score=True)

# Fit the random search model
rf_random.fit(X_train, y_train)
rf_random.best_score_

# accuracy
print('accuracy best score:')
print(rf_random.best_score_)
print('best model parameters:')
print(rf_random.best_params_)

Fitting 5 folds for each of 200 candidates, totalling 1000 fits
accuracy best score:
0.9385597550768688
best model parameters:
{'n_estimators': 800, 'min_samples_split': 2, 'min_samples_leaf': 1, 'max_features': 'auto', 'max_depth': 100, 'bootstrap': True}


# model performace after finding the best model
best_random = rf_random.best_estimator_
best_random.fit(X_train, y_train)
predictions_score = best_random.predict_proba(X_test)[:,1]
predictions= best_random.predict(X_test)

CM = confusion_matrix(y_test.to_numpy(), predictions)#,labels=[0,1]
print(classification_report(y_test.to_numpy(), predictions))
# Visualize it as a heatmap
sns.heatmap(CM, fmt = 'd' , cmap="YlGnBu", cbar = False,  annot=True)
plt.show()

              precision    recall  f1-score   support

           0       0.95      0.99      0.97       708
           1       0.93      0.68      0.79       126

    accuracy                           0.94       834
   macro avg       0.94      0.84      0.88       834
weighted avg       0.94      0.94      0.94       834


import sklearn.metrics as metrics
fpr, tpr, threshold = metrics.roc_curve(y_test, predictions_score)
roc_auc = metrics.auc(fpr, tpr)
plt.title('Receiver Operating Characteristic')
plt.plot(fpr, tpr, 'b', label = 'AUC = %0.2f' % roc_auc)
plt.legend(loc = 'lower right')
plt.plot([0, 1], [0, 1],'r--')
plt.xlim([0, 1])
plt.ylim([0, 1])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
plt.show()

#High TPR/Recall is achieved while maintaining a low FPR. Depending on our tolerance for False Positives, we can increase our TPR as well by shifting the threshold.


# XGBoost:
# Finding best XGB model Model with random grid search


#findng best hyperparameters

plt.rcParams.update({'figure.figsize': (12.0, 8.0)})
plt.rcParams.update({'font.size': 14})


model = XGBClassifier()

param_grid = {
        'min_child_weight': [1, 5, 10],
        'gamma': [0.5, 1, 1.5, 2, 5],
        'subsample': [0.6, 0.8, 1.0],
        'colsample_bytree': [0.6, 0.8, 1.0],
        'max_depth': [3, 4, 5]
}

gs = GridSearchCV(
    estimator=model,
    param_grid=param_grid, 
    cv=5, 
    n_jobs=-1, 
    scoring='precision',
    verbose=2
)
xgb_grid = gs.fit(X_train, y_train)
    
# accuracy
print('best score:')
print(xgb_grid.best_score_)
print('best model parameters:')
print(xgb_grid.best_params_)

Fitting 5 folds for each of 405 candidates, totalling 2025 fits
best score:
0.944438682496276
best model parameters:
{'colsample_bytree': 0.8, 'gamma': 5, 'max_depth': 4, 'min_child_weight': 5, 'subsample': 1.0}


best_XGB = xgb_grid.best_estimator_
best_XGB.fit(X_train, y_train)
predictions_score = best_XGB.predict_proba(X_test)[:,1]
base_predictions = best_XGB.predict(X_test)

CM = confusion_matrix(y_test.to_numpy(), base_predictions)
print(classification_report(y_test.to_numpy(), base_predictions))
# Visualize it as a heatmap
sns.heatmap(CM, fmt = 'd' , cmap="YlGnBu", cbar = False,  annot=True)
plt.show()

              precision    recall  f1-score   support

           0       0.95      0.99      0.97       708
           1       0.90      0.71      0.80       126

    accuracy                           0.94       834
   macro avg       0.93      0.85      0.88       834
weighted avg       0.94      0.94      0.94       834


import sklearn.metrics as metrics
fpr, tpr, threshold = metrics.roc_curve(y_test, predictions_score)
roc_auc = metrics.auc(fpr, tpr)
plt.title('Receiver Operating Characteristic')
plt.plot(fpr, tpr, 'b', label = 'AUC = %0.2f' % roc_auc)
plt.legend(loc = 'lower right')
plt.plot([0, 1], [0, 1],'r--')
plt.xlim([0, 1])
plt.ylim([0, 1])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
plt.show()

##High TPR/Recall is achieved while maintaining a low FPR. Depending on our tolerance for False Positives, we can increase our TPR as well by shifting the threshold.


# features mean impact on XGBoost model output
shap_values = shap.TreeExplainer(best_XGB).shap_values(X_train)
shap.summary_plot(shap_values, X_train.columns, plot_type="bar")

C:\Users\ElHarameinSamyJaby\Anaconda3\lib\site-packages\xgboost\core.py:90: UserWarning:

ntree_limit is deprecated, use `iteration_range` or model slicing instead.


# features VALUE impact on XGBoost model output
#Product International high value (True) has the biggest impact on model output.
#The same goes for service call, eve minutes and churn rate.
#Product voice call low values (False) has the biggest impact on model output.
#Total, EUR feature in XGBoost has less impact on the model and random forest.

shap.summary_plot(shap_values, X_train)


#Random Forest:
#Accuracy best score: 0.93

#XGBoost:
#Accuracy best score: 0.94

#The focus for the Random search is to find the best parameters to maximize precision average.

#Confusion matrix for our Test predictions shows the values as follows:
#Predicted not to churn and not churned TN    |  Predicted not Churned but churned FP
#Predicted to churn and not churned FN          | Predicted to churn and churned TP
#The main focus should be on minimizing false positive and false negative, this can be done by optimizing the precision and recall.
#For our demonstration I focused on optimizing the precision.


#XGBoost showed slightly better score and false negative, which is our focus to minimize the number of users which we predict as positive which eventually won’t churn.



#Some of the assumptions were correct regarding the features, since service call, total EUR, produce: international were the most important features for RF.


#there are differnt types of testing which can be considered 

#-Cross-validation while training the model: for this scenario 5 fold cross validation is used to test the trained model

#-Automated tests for model verification: by applying smoke test, unit tests and finally integration tests
# unit tests and integration tests help to prevent low-quality data from entering the training pipeline.
#-Manual model evaluation and validation: done by domain expert generated and reviewed a model quality report


#.ppt attached


#The following steps will be taken in case i have access to Azure synapse:

#-Open Azure machine learning workspace lined service in synapse
#-Open Azure new notebook for EDA and preprocessing
#-Access to the churn data maybe in sql in synapse workspace
#-Make the EDA and preprocessing using synapse workspace
#-Train and test the needed models in the created notebook, each model in a separate file, analyse the results 
#-Use one of ml model interpretation tools to share the results, ex. SHAP, Mimic 

#The following steps will be taken to automate the model run:
#-Set a threshold for the new model score generated, for example 0.85
#-The threshold will guarantee the quality of the regularly trained model
#-The score here is either precision or recall in case of churn
#-Add the init function to deploy the model
#-Add the run function to predict unseen new data
#-Use SHAP to explain and monitor the model performace

	customerID	gender	US State	Duration	Phone Code	Telephone Number	Product: International	Product: Voice mail	vmail	...	eve EUR	night minutes	night calls	night EUR	international minutes	international calls	internatonal EUR	service calls	PaperlessBilling	churn
0	7590-VHVEG	Female	KS	128	415	382-4657	no	yes	25	...	16.78	244.7	91	11.01	10.0	3	2.70	1	Yes	False
1	5575-GNVDE	Male	OH	107	415	371-7191	no	yes	26	...	16.62	254.4	103	11.45	13.7	3	3.70	1	No	False
2	3668-QPYBK	Male	NJ	137	415	358-1921	no	no	0	...	10.30	162.6	104	7.32	12.2	5	3.29	0	Yes	False
3	7795-CFOCW	Male	OH	84	408	375-9999	yes	no	0	...	5.26	196.9	89	8.86	6.6	7	1.78	2	No	False
4	9237-HQITU	Female	OK	75	415	330-6626	yes	no	0	...	12.61	186.9	121	8.41	10.1	3	2.73	3	Yes	False

	SeniorCitizen	Duration	Phone Code	vmail	Call day minutes	total day calls	Total, EUR	eve minutes	eve calls	eve EUR	night minutes	night calls	night EUR	international minutes	international calls	internatonal EUR	service calls
count	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000	3333.000000
mean	0.157516	101.064806	437.182418	8.099010	179.775098	100.435644	30.562307	200.980348	100.114311	17.083540	200.872037	100.107711	9.039325	10.237294	4.479448	2.764581	1.562856
std	0.364341	39.822106	42.371290	13.688365	54.467389	20.069084	9.259435	50.713844	19.922625	4.310668	50.573847	19.568609	2.275873	2.791840	2.461214	0.753773	1.315491
min	0.000000	1.000000	408.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	23.200000	33.000000	1.040000	0.000000	0.000000	0.000000	0.000000
25%	0.000000	74.000000	408.000000	0.000000	143.700000	87.000000	24.430000	166.600000	87.000000	14.160000	167.000000	87.000000	7.520000	8.500000	3.000000	2.300000	1.000000
50%	0.000000	101.000000	415.000000	0.000000	179.400000	101.000000	30.500000	201.400000	100.000000	17.120000	201.200000	100.000000	9.050000	10.300000	4.000000	2.780000	1.000000
75%	0.000000	127.000000	510.000000	20.000000	216.400000	114.000000	36.790000	235.300000	114.000000	20.000000	235.300000	113.000000	10.590000	12.100000	6.000000	3.270000	2.000000
max	1.000000	243.000000	510.000000	51.000000	350.800000	165.000000	59.640000	363.700000	170.000000	30.910000	395.000000	175.000000	17.770000	20.000000	20.000000	5.400000	9.000000

	international minutes	eve calls	total day calls	US State
0	10.0	99	110	KS
1	13.7	103	123	OH
2	12.2	110	114	NJ
3	6.6	88	71	OH
4	10.1	122	113	OK
...	...	...	...	...
3328	9.9	126	77	AZ
3329	9.6	55	57	WV
3330	14.1	58	109	RI
3331	5.0	84	105	CT
3332	13.7	82	113	TN

	US State	total day calls rate	night calls rate	international calls rate	Duration rate	Total, EUR rate	service calls rate	Call day minutes rate	night minutes rate	churn rate
0	AK	95.269231	97.346154	4.807692	97.192308	30.325769	1.557692	178.384615	192.326923	0.057692
1	AL	99.100000	98.612500	4.700000	98.025000	31.621500	1.562500	186.010000	187.285000	0.100000
2	AR	102.145455	99.054545	4.690909	96.981818	29.940000	1.981818	176.116364	205.454545	0.200000
3	AZ	100.109375	99.984375	4.859375	100.468750	29.173125	1.593750	171.604688	194.004687	0.062500
4	CA	104.500000	99.205882	4.441176	99.235294	31.206176	1.470588	183.564706	198.508824	0.264706
5	CO	95.121212	97.060606	4.106061	102.545455	30.381818	1.787879	178.712121	189.898485	0.136364
6	CT	98.432432	102.283784	4.108108	99.040541	29.774324	1.513514	175.140541	205.997297	0.162162
7	DC	100.166667	97.666667	3.907407	105.722222	29.135370	1.351852	171.379630	206.348148	0.092593
8	DE	98.639344	101.213115	4.295082	101.918033	29.680164	1.622951	174.583607	203.900000	0.147541
9	FL	101.000000	100.333333	3.984127	109.571429	30.521429	1.571429	179.533333	196.147619	0.126984
10	GA	103.055556	105.425926	4.055556	103.592593	31.455185	1.925926	185.025926	193.746296	0.148148
11	HI	102.811321	99.150943	4.622642	105.471698	29.913019	1.547170	175.962264	203.713208	0.056604
12	IA	103.045455	100.136364	4.727273	94.318182	30.195455	1.590909	177.613636	191.490909	0.068182
13	ID	106.219178	100.424658	4.561644	104.493151	30.365753	1.671233	178.619178	202.595890	0.123288
14	IL	101.689655	99.224138	4.344828	100.517241	29.511552	1.137931	173.591379	197.605172	0.086207
15	IN	107.056338	101.929577	4.281690	100.183099	33.410000	1.690141	196.525352	210.242254	0.126761
16	KS	101.028571	99.028571	4.500000	106.785714	32.565143	1.385714	191.555714	203.970000	0.185714
17	KY	100.338983	101.694915	4.322034	91.830508	29.539153	1.711864	173.754237	198.355932	0.135593
18	LA	103.509804	99.588235	4.647059	108.235294	30.324706	1.568627	178.376471	201.396078	0.078431
19	MA	102.292308	96.600000	4.446154	91.861538	30.617538	1.215385	180.103077	204.007692	0.169231
20	MD	99.357143	95.757143	4.428571	105.271429	33.529571	1.700000	197.228571	198.614286	0.242857
21	ME	99.500000	98.967742	4.596774	97.629032	31.495806	1.741935	185.262903	198.833871	0.209677
22	MI	101.205479	102.630137	4.547945	98.109589	30.700959	1.630137	180.593151	192.657534	0.219178
23	MN	99.119048	102.130952	4.333333	101.571429	31.170952	1.547619	183.354762	209.680952	0.178571
24	MO	99.936508	101.269841	4.857143	98.619048	28.987302	1.555556	170.506349	209.146032	0.111111
25	MS	100.707692	100.892308	4.523077	99.907692	30.248462	1.692308	177.929231	200.996923	0.215385
26	MT	95.220588	102.382353	4.544118	92.676471	29.581618	1.632353	174.007353	197.205882	0.205882
27	NC	102.794118	100.500000	4.117647	104.779412	31.475294	1.558824	185.145588	197.144118	0.161765
28	ND	100.145161	102.274194	4.919355	106.209677	31.848548	1.451613	187.338710	199.796774	0.096774
29	NE	97.409836	101.639344	4.213115	98.721311	30.169344	1.459016	177.465574	206.427869	0.081967
30	NH	98.964286	96.678571	4.714286	101.875000	30.146250	1.517857	177.328571	208.614286	0.160714
31	NJ	100.102941	100.147059	4.691176	100.205882	33.358676	1.676471	196.225000	206.383824	0.264706
32	NM	102.241935	100.870968	4.838710	104.370968	29.142903	1.596774	171.429032	200.193548	0.096774
33	NV	101.424242	101.363636	4.333333	102.803030	29.992879	1.590909	176.425758	208.645455	0.212121
34	NY	98.240964	99.409639	4.638554	100.698795	29.770361	1.710843	175.114458	203.268675	0.180723
35	OH	99.628205	103.256410	4.371795	102.153846	31.157564	1.358974	183.274359	204.491026	0.128205
36	OK	100.262295	100.983607	4.803279	108.262295	30.584754	1.786885	179.909836	196.947541	0.147541
37	OR	95.730769	102.679487	4.333333	101.166667	29.962821	1.730769	176.246154	199.925641	0.141026
38	PA	98.644444	98.911111	3.866667	104.044444	32.024444	1.266667	188.375556	195.864444	0.177778
39	RI	100.430769	100.338462	4.646154	101.907692	28.472154	1.415385	167.478462	204.052308	0.092308
40	SC	102.466667	100.183333	4.183333	97.016667	28.295833	1.583333	166.441667	195.136667	0.233333
41	SD	94.816667	94.016667	4.483333	105.450000	32.247667	1.400000	189.690000	201.310000	0.133333
42	TN	101.452830	94.264151	4.339623	92.716981	29.881698	1.320755	175.771698	210.426415	0.094340
43	TX	99.986111	101.041667	4.625000	97.541667	30.858472	1.611111	181.516667	195.288889	0.250000
44	UT	98.958333	100.625000	4.625000	100.958333	31.207222	1.375000	183.569444	190.519444	0.138889
45	VA	105.363636	94.922078	4.740260	105.935065	30.131688	1.597403	177.244156	212.963636	0.064935
46	VT	100.328767	100.767123	4.602740	104.424658	30.946301	1.739726	182.031507	206.989041	0.109589
47	WA	101.196970	102.909091	4.454545	102.060606	30.386667	1.530303	178.742424	200.045455	0.212121
48	WI	96.679487	100.461538	4.128205	98.884615	30.452308	1.423077	179.130769	199.229487	0.089744
49	WV	103.783019	100.566038	4.415094	95.433962	29.572358	1.500000	173.950943	201.055660	0.094340
50	WY	102.220779	102.025974	4.974026	105.740260	30.628961	1.441558	180.170130	199.167532	0.116883

	customerID	gender	SeniorCitizen	US State	Duration	Phone Code	Telephone Number	Product: International	Product: Voice mail	vmail	...	total day calls rate	night calls rate	international calls rate	Duration rate	Total, EUR rate	service calls rate	Call day minutes rate	night minutes rate	churn rate	AreaCode
0	7590-VHVEG	1	0	KS	128	415	382-4657	0	1	25	...	1.088801	0.918927	0.666667	1.198662	1.383995	0.721649	1.383932	1.199686	0.185714	382
1	3712-PKXZA	0	0	KS	70	408	411-4582	0	0	0	...	1.207579	1.130987	0.000000	0.655518	1.211725	2.164948	1.211658	0.986420	0.185714	411
2	4959-JOSRX	1	0	KS	92	408	386-4170	1	0	0	...	1.098699	0.807848	0.444444	0.861538	0.326730	0.721649	0.326798	1.086925	0.185714	386
3	5583-SXDAG	0	0	KS	137	408	352-8202	0	0	0	...	0.871041	0.848240	0.888889	1.282943	1.074769	1.443299	1.074883	1.421287	0.185714	352
4	9571-EDEBV	0	0	KS	132	415	369-3214	0	0	0	...	1.088801	1.151183	1.111111	1.236120	0.435435	0.721649	0.435382	0.719223	0.185714	369
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3328	5762-TJXGK	1	0	ND	84	408	351-1894	0	1	38	...	1.058464	0.850654	1.016393	0.790888	1.030188	1.377778	1.030220	1.303324	0.096774	351
3329	2122-SZZZD	0	0	ND	131	408	393-9548	0	1	33	...	0.998550	1.212427	1.016393	1.233409	0.945412	0.688889	0.945347	1.268289	0.096774	393
3330	3889-VWBID	0	0	ND	75	408	396-4171	0	1	24	...	1.188275	1.036430	0.406557	0.706150	1.203823	2.066667	1.203702	1.355878	0.096774	396
3331	5150-ITWWB	0	1	ND	27	415	405-1589	0	0	0	...	0.669029	1.065762	1.219672	0.254214	1.213870	0.688889	1.213844	0.307312	0.096774	405
3332	0587-DMGBH	1	0	ND	148	415	396-4234	1	0	0	...	1.108391	0.606214	1.016393	1.393470	1.167400	0.688889	1.167404	1.388411	0.096774	396

ITC Client Churn¶

1.Analyze the data, clean the data, transform if needed. Provide a few key visuals.¶

2. Prepare a graph that shows the distribution of total number of minutes for international calls geographically in breakdown¶

3. Make your assumptions on the relevant features for churn prediction. What extra features would you like to have, as you believe, they are crucial to explain churn.¶

4. Generate new features that are relevant to explain churn. For example: number of call per day for a particular person as compared to number of calls for a comparable sample (i.e. for the same state).¶

5. Select a model, justify your choice, build a model to predict churn. (Auto ML shall not be used)¶

6. Explain your model results, explain which features are important, which are not. Were your assumptions correct? If not, why do you think that is the case?¶

7. How would you test the model results? What is the best practice?¶

8. Present the results (showcase the model, explain each step, if needed support your presentation with power point slides) and insights that you managed to extract out of the data.¶

9. What if you had an access to Azure synapse workspace, how would you approach the task? How would you automate the model run?¶

	count	unique	top	freq
customerID	3333	3333	7590-VHVEG	1
gender	3333	2	Male	1705
SeniorCitizen	3333	2	0	2808
US State	3333	51	WV	106
Duration	3333	212	105	43
Phone Code	3333	3	415	1655
Telephone Number	3333	3306	396-8265	4
Product: International	3333	2	no	3010
Product: Voice mail	3333	2	no	2411
vmail	3333	46	0	2411
Call day minutes	3333.0	1667.0	154.0	8.0
total day calls	3333	119	102	78
Total, EUR	3333.0	1667.0	26.18	8.0
eve minutes	3333.0	1611.0	169.9	9.0
eve calls	3333	123	105	80
eve EUR	3333.0	1440.0	14.25	11.0
night minutes	3333.0	1591.0	191.4	8.0
night calls	3333	120	105	84
night EUR	3333.0	933.0	9.66	15.0
international minutes	3333.0	162.0	10.0	62.0
international calls	3333	21	3	668
internatonal EUR	3333.0	162.0	2.7	62.0
service calls	3333	10	1	1181
PaperlessBilling	3333	2	Yes	1959
churn	3333	2	False	2850

	customerID	gender	SeniorCitizen	US State	Duration	Phone Code	Telephone Number	Product: International	Product: Voice mail	vmail	...	total day calls rate	night calls rate	international calls rate	Duration rate	Total, EUR rate	service calls rate	Call day minutes rate	night minutes rate	churn rate	AreaCode
0	2549	1	0	16	126	1	1908	0	1	19	...	1449	795	120	1828	2932	43	2932	2561	35	55
1	1199	0	0	16	68	0	2944	0	0	0	...	1828	1566	0	553	2492	145	2492	1569	35	84
2	1641	1	0	16	90	0	2054	1	0	0	...	1484	437	73	1021	45	43	45	2066	35	59
3	1868	0	0	16	135	0	905	0	0	0	...	660	555	168	1997	1946	93	1946	3133	35	25
4	3201	0	0	16	130	1	1475	0	0	0	...	1449	1637	216	1917	103	43	103	432	35	42
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3328	1929	1	0	28	82	0	838	0	1	32	...	1334	563	191	839	1758	88	1758	2896	12	24
3329	669	0	0	28	129	0	2311	0	1	27	...	1112	1830	191	1908	1399	38	1399	2789	12	66
3330	1261	0	0	28	73	0	2397	0	1	18	...	1781	1219	50	653	2463	139	2463	3025	12	69
3331	1708	0	1	28	25	1	2702	0	0	0	...	151	1328	237	91	2502	38	2502	11	12	78
3332	197	1	0	28	146	1	2398	1	0	0	...	1511	61	191	2219	2327	38	2327	3085	12	69